Yves here. Admittedly, the AI bubble may pop before its large and ginormous energy demands really kick in. But (as occasional past posts help attest), the US grid was already in creaky shape. Trying to push more power through it without a major upgrade/redo is asking for serious trouble.
The post below is a testament to the lack of leadership in the US. both the failure to question the necessity/desirability of testing the electric grid to (at least near) destruction, and deer-in-the-headlights quality of responses given that no effort seems to be underway to reduce AI demand (how about taxes?!?!). You’ll notice, in keeping, that this article takes the whistling-past-the-graveyard view that maybe this problem will sort itself out.
By Robert Rapier, a chemical engineer in the energy industry. He has 25 years of international engineering experience in the chemical, oil and gas, and renewable energy industries, and holds several patents related to his work. Originally published at OilPrice
- The US power grid is under significant stress due to rapidly increasing electricity demand from AI data centers and electric vehicles, coupled with the retirement of traditional power plants.
- The grid faces mounting vulnerabilities from extreme weather events, cybersecurity threats, and physical sabotage, while policy gridlock and infrastructure delays hinder necessary upgrades.
- Despite challenges, the situation presents investment opportunities in grid modernization, diversified generation, and solutions for energy storage and load management.
America’s power grid is straining under the weight of a fast-changing energy landscape. Beyond the usual summer hum of air conditioners, power demand is surging from electric vehicle chargers and sprawling new data centers. At the same time, the infrastructure built to deliver reliable electricity is aging and showing its limits. From Texas heatwaves to California blackouts, the warning signs are impossible to ignore.
This isn’t a technical challenge—it’s an economic and political reckoning. If the grid fails, it won’t be because we lacked solutions. It will be because we didn’t act quickly enough.
A Demand Surge Few Anticipated
For nearly two decades, U.S. electricity demand was flat. Now, consumption is climbing, driven by technologies that arrived faster than planners expected.
Artificial intelligence has unleashed a wave of data center construction. These facilities, dense with high-performance servers and cooling systems, are among the most power-hungry assets in the country. In 2023, AI data centers consumed about 4.4% of U.S. electricity, and that share could triple by 2028, according to Penn State’s Institute of Energy and the Environment.
Northern Virginia—“Data Center Alley”—now handles 70% of global internet traffic, pushing utilities like Dominion Energy to scramble for capacity. Meanwhile, Microsoft and Google warn that a shortage of skilled electricians could delay expansion, with estimates that the U.S. will need 500,000 more electricians in the next decade.
Electric vehicles, heat pumps, and electrified industry are adding further strain. The Pacific Northwest Utilities Conference Committee projects growth equivalent to seven Seattle-sized cities within ten years. The Energy Information Administration (EIA) expects U.S. electricity sales to rise from 4,097 billion kWh in 2024 to 4,193 billion kWh in 2025, with similar gains to follow.
And finally, there’s climate. As extreme heat events multiply, cooling demand in places like Texas and Arizona is surging—driving peak loads to new records.
The Supply Gap: Retirements Outpace Replacement
Just as demand is accelerating, the U.S. is retiring some of its most dependable sources of power.
The EIA projects 12.3 gigawatts (GW) of capacity will retire in 2025, a 65% jump over 2024. That includes 8.1 GW of coal, such as the 1,800-MW Intermountain Power Project in Utah, plus another 2.6 GW of natural gas. These plants provide round-the-clock power that intermittent sources cannot yet replace.
Wind and solar capacity continue to grow, but not fast enough. The Department of Energy’s July 2025 Resource Adequacy Report warns that only 22 GW of firm generation is expected by 2030—well short of the 104 GW needed for peak demand. Transmission bottlenecks, permitting delays, and slow adoption of long-duration storage compound the problem.
Grid operators from PJM, MISO, ERCOT, and others told Congress bluntly in March 2025: “Demand is accelerating, supply is lagging, and current tools may not be enough to bridge the gap.”
Growing Vulnerabilities: Weather, Cyber, and Sabotage
Beyond the supply-demand imbalance, the grid faces mounting risks.
Heatwaves, wildfires, and storms are stressing transmission systems nationwide. Events once considered rare—like the 2003 Northeast blackout that hit 50 million people—are now seen as precursors of larger disruptions.
As smart grids and distributed resources proliferate, so do digital entry points for hackers. In 2024, DOE funded 16 cybersecurity projects, including Georgia Tech’s AI-based “DerGuard” system to monitor risks in distributed energy.
Sabotage at substations and transmission lines is also rising. With more than 160,000 miles of high-voltage lines and 7,300 plants, much of it decades old, the system is a sprawling, exposed target. Homeland Security now classifies grid protection alongside nuclear and water infrastructure.
Policy Gridlock and Infrastructure Delays
Despite the alarms, policy responses remain sluggish.
Jurisdiction is a big part of the problem. Regional transmission operators manage the grid but don’t own generation or lines. Utilities do, while states control siting and permitting. The result is a patchwork that slows progress.
As of mid-2024, transmission projects across the U.S. faced delays of five to seven years due to permitting hurdles, interconnection bottlenecks, and supply chain constraints. By mid-2025, lead times for large power transformers had stretched beyond 30 months, with some units requiring up to four years for delivery—posing serious risks to grid reliability and expansion.
Even bipartisan efforts like the CIRCUIT Act—introduced in February 2025 to incentivize domestic transformer production through a 10% tax credit—remain stalled in committee, despite widespread industry support and urgent supply chain concerns.
Meanwhile, subsidies continue to favor intermittent renewables over firm capacity. The Inflation Reduction Act accelerated clean energy deployment, but without parallel investment in balancing technologies, reliability risks grow.
What’s Being Done
Federal and private efforts are ramping up, though often as short-term fixes.
DOE has delayed retirements of coal and gas plants and issued reliability directives under the Federal Power Act. These measures keep the lights on but do little to build long-term resilience. DOE also aims to increase long-distance capacity 16% by 2030, adding 7,500 miles of new lines. But permitting delays and local resistance remain obstacles.
In 2025, DOE launched $32 million in pilot projects for smart EV charging, responsive buildings, and distributed energy integration. These solutions could eventually scale, but utilities and regulators must buy in.
Investor Implications: Reliability as a Premium
For investors, grid instability is a risk, but also an opportunity.
Companies like NextEra Energy, Dominion, and Avangrid are investing billions in grid modernization and diversified generation. Avangrid alone plans $20 billion through 2030 across 23 states.
Independent power producers are also benefiting from the shifting landscape. NRG Energy, one of the nation’s largest competitive power suppliers, has seen its shares climb sharply as rising demand boosts wholesale electricity prices. Unlike regulated utilities, NRG and its peers compete in deregulated markets, where higher load growth and tighter capacity directly translate into stronger margins. That dynamic could make competitive generators an overlooked winner in a strained grid environment.
Firms like Fluence, Stem Inc., and Tesla Energy are seeing growing demand for storage and microgrid solutions. Pilot programs backed by DOE may open new markets for software-driven load management.
With coal exiting and renewables constrained, power generated by nuclear energy and natural gas retain a “reliability premium.” Deloitte estimates the U.S. power sector will need $1.4 trillion in new capital between 2025 and 2030, with similar levels required through 2050. Firms able to supply firm generation or grid services stand to benefit.
Conclusion: Crisis or Course Correction?
The U.S. power grid isn’t collapsing—but it is under pressure like never before. Demand growth, baseload retirements, extreme weather, and policy paralysis are colliding to create a fragile system.
Whether this moment becomes a crisis or a correction depends on how quickly policymakers, utilities, and investors adapt. The tools exist—firm generation, smart load management, and modern transmission. But without faster coordination and realistic incentives, the U.S. risks trading energy abundance for energy fragility.
This is a messy article. Isn’t it?. Tries to pitch too many points without substantial information supporting much of it but lots of randomly scattered data. Few days ago I read there are now significant investments in new NG-based utility generation but this is omitted in the article.
For what i have recently learnt, many and possibly the largest AI centres are relatively concentrated in hubs in Northern Virginia, Texas and few other places so if this creates constraints these are to be addressed in relatively few spots and the infrastructure being built will have little impact in the capacity of most of the grid. In contrast EV chargers and other electrification efforts is much more widely distributed and will require its own investments, specially in the distribution network. Addressing AI does not solve EV.
In any case I very much agree that the grid would require centralised planning, tight coordination and a sea change in jurisdiction rules.
Yes, its a messy article, and also ignores that the US is in reality three to four (depending on how you count) largely separate grids, with a number of sub-regional grids.
As we’ve seen in Ukraine and Syria, even under extreme strain older grids don’t ‘collapse’, they have lots of inbuilt resilience – they become unstable and unreliable, but this is a gradual and unpredictable process.
As I’ve stated before, I consider the impact of AI and data centres to be over-stated as a problem. If they cause overload in a local grid that’s not an infrastructural problem – its a contractual one. Most heavy commercial users will have outages built into the contract. Quite simply, if they demand more than the grid can supply, they get cut off, simple as that. This is why most of those centres have built in generating back up (which itself is an environmental problem, but a separate one). Its not absolute load demand that causes problems for grid, its how the peaks and troughs are managed. Grid management is an extremely complicated issue as it involves the interplay of supply, demand, economic costs, contractual and legal obligations, risk assessment, individual circuit engineering capacity, future projections of costs, demand with all the uncertainties involved, and so on.
The US is facing a familiar problem of all mature economies in that it has developed a habit of free riding on past waves of development. The political glamour of upgrading and extending electricity supply has long gone and the requirements now are for boring incremental upgrades on long investment cycles, something politicians are notoriously loathe to spend any time thinking about. A key problem of course is that economic and technological changes impacting on grids often happen on a shorter cycle than grid planning and implementation. In reality, the engineering necessity is to significantly overbuild, and accept some level of wasted investment in order to have the necessary resilience. This is a hard sell when there are many other demands on investment.
So far the most significant impact of AI is that it is driving up electricity bills specially in the non-industrial sector: commercial, residential etc. this, IMO, suggests that the regulatory framework and the pricing schemes should also be revised. The default approach in, as you say, mature economies leads to a few profiting while everyone else pays for it. It is also an indirect subsidy to those commercial projects.
Messy is being generous. Another classic from oil price.
Agreed that the market pricing structures need to be and I believe are starting to change. Just look at the pricing differences between co-ops and IOU’s.
As to the power plant retirement plans , which many were put into place under the Biden administration under completely irrational reasons, most won’t happen. You can’t shut down a large % of power without first having their replacements on line. I mean you can but it’s going to set up big problems.
And yes the country as has been explained didn’t really grow electricity from about 2000-2020.
Some regional grids did see the change coming and moved forward with improvements.
The more urbanized areas it’s much harder to install new powerlines and plants thinking PJM.
As to transformers and gas turbines, and other grid components, they are available from China among other places at a much faster time frame than US suppliers. I mean probably getting rid of the 50%!steel tariffs would be a simple place to start.
Take geothermal. The main issue is govt regulations that make it take way too long. 100-1000’s of GW can come on line pretty quickly if they can get permits in under 15 years. That’s a real number. I was at a talk on geothermal and one speaker is an actual developer. 15 years is what it took his last well for permits. While a mining company took weeks to get a well permit. Or oil and gas guys probably takes seconds.
Here is to Hope.
Aye, and it’s not just geothermal where permitting takes excessively long:
High-voltage transmission lines are another.
https://reason.com/2023/04/17/it-took-15-years-for-the-feds-to-approve-a-700-mile-electric-line/
Plus pumped storage projects. I’ve been following the 1.3GW Eagle Mountain project in California, which would increase the state’s grid ~30GWh of grid storage capability by a substantial 22GWh. It would really help with their curtailment issues. The permitting process started in 2008, and they’re expecting to break ground in 2027, 19 years later. Presuming no additional delays, of course.
https://www.waterboards.ca.gov/waterrights/water_issues/programs/water_quality_cert/eaglemtn_ferc13123.html
https://www.power-technology.com/marketdata/power-plant-profile-eagle-mountain-us/
Yup
And
Sun Zia transmission line some 12 years in the permitting, is being built.
NM wind power to Phoenix.
Which are the explanations to those delays you and Grumpy Engineer mention? Environmental issues or other types of complexities?— I feel sorry with this question having not read the links provided by G.E.
It’s mainly environmental reviews (my second link captures stuff from just the California Water Board; other state- and federally-required items are not listed).
But then there are also the lawsuits. Infrastructure projects can run afoul of historical/cultural concerns and bring out the lawyers. For a recent example, see https://gilaherald.com/federal-court-rules-in-favor-of-tribes-against-sunzia/. If SunZia has to re-route in the middle of the project, there’s no way they’ll finish on time or on budget.
Thank you!
My view is that Matt Stoller explained the situation much better than the conventional wisdom that AI is the driving force behind grid issues and rising prices. The neoliberal shift to utilities as profit earning, shareholder serving (self-serving for Execs) entities with monopolies that prioritize share prices over quality service explains it all.
https://www.thebignewsletter.com/p/data-centers-arent-the-main-villain
Thanks for this. It’s a great reminder of how well Matt Stoller explains why we can’t have nice things.
Agreed. The nation was sold the idea that deregulation and decoupling generation and delivery would create competitive markets and that would drive investment in building out generation as needed. However, there is nothing in any part of electricity generation or delivery that does not include massive, fixed costs which means every part of it operates on a downward sloping cost curve or by definition is monopolistic in its cost and pricing structure. The limiting of supply is how profits are maximized by this type of entity. There is no private utility incentive to invest in optimizing any part of the grid.
Experts do a lousy job of explaining it to Normies,
Buying-selling-moving electrons is not like the market for bread
A suboptimal, but least bad, stopgap solution is easy: with smart grid tech, it’s easy to ID data centers or EV chargers. Just triage the grid and cut off the centers and EV chargers (not many steel mills in No. VA to cut off, lol) during potential brownouts
that’ll never happen given the lobbying power of Tech
Any planned new generation capacity is unlikely to be completed before the AI power demand reduces due to either usage drop (current demand bubble) or more efficient processing/code.
“Northern Virginia—“Data Center Alley”—now handles 70% of global internet traffic.”
Imagine a small nuke setting off an EMP over this concentration of data centers. In the late fifties with the Cold War in mind, air traffic control centers were purposely located outside of major cities. No such forethought today.
One major thing everyone is missing is that the physical grid is decaying and falling apart. The local power thieves where I live charge $25 dollars a month connection fee to maintain the infrastructure. When a member of the COOP have been connected for 20 years the get some sort of rebate. Well the management wanted to increase their bonus, so the got in to the reserve and gave the members a $20-100 rebate. The power line are in bad shape in a lot of areas. Here in the Northwest we’ve had numerous major fires started by power line failures. The Obama administration gave the power companies billions to upgrade the grid. All that happened is they put in smart meters, laid off meter reader and linemen, fixed little or nothing and management got big bonuses. The clown here about four years ago got rid of all the linemen and went to contractor services. A year later we had a massive winter storm took out a lot of power lines.
All the contractor service companies were all engaged in the Southwest in hurricane damage repair and some customers went a month without grid power. They did hire a few linemen back. The power lines that run a long my property are 100+ years old. At least their copper. The grid aluminium lines are getting in bad shape especially the major transmission lines. Aluminium line start crystallizing after a period of time. This increases the resistive power loss, that increases heat in the power line causing the lines to to stretch and sag. There’s no national plan to even upgrade the major transmission lines. How in hell are the going to power it? You can build any type power generation system but need to have a way to transmit the power. So most of the AI cost will end up on the rate payers. The local power rates here have gone up at least 10% per year this year it was 12%. The local COOP byes most of it’s power from Bonneville Power Administration. Four years ago to get power to my property was $35K. So I’m 100% off the grid, with backup generator and a rack Lion batteries for a lot less than $35K. By the way I have hit ROI. My power is basically free except when I need to run the backup generator. Also, the local power COOP will not buy my excess power. I offered to give it to them for free if they would bank say 10% of the power and let me use it instead of my generator. No luck. In the summer I have enough spare to cover the AC for at 2 houses. So we as country have no nation wide energy plan, like China, no one even looks at reality. The F**ing government will stick us with cost of AI infrastructure while most likely a people will not be able to afford power and food at the same time. Where in the hell are we going?
What people are not taking into account is the incredible drop in price for battery storage.
The issue most of the grid has is peak not daily.
More storage can all but eliminate the more urgent calls to provide peak load power
Battery storage is about $.01-$.02 per kWh. It’s cheap
It’s the fastest to install, small footprint and modular.
Lots of ways to work the system to make it work way better.
I did notice that the article strongly referenced DOE as part of the solution – but ignores Trumpian cuts to the Department of Energy…… Hmmm, whistling indeed!
https://thefulcrum.us/governance-legislation/department-of-energy-under-trump
If a third of the American population dies over 2-3 years as the result of Covid caused immune dysregulation demand will be reduced significantly.
Gotta look on the bright side!
I have fixed your doorbell for the ringing. There is no charge.
Peter Sellers as Inspector Clouseau